2.1 Patient population
This retrospective study was approved by the institutional review board and the informed consent was waived. Patients with EC who underwent 3.0 MRI before treatment in our institution from January 2015 to December 2019 were included retrospectively. Inclusion criteria are as follows: (a) Pathologically confirmed EC; (b) available clinical and postoperative pathological data; (c) MRI was performed within 1 month before surgery; (d) surgical treatment and follow-up were performed in our hospital. Exclusion criteria are as follows: (a) diagnosis of malignant tumor other than endometrial cancer (n = 4); (b) distant metastasis occurred at the time of diagnosis (IVB stage) (n = 5); (c) tumors were invisible on MRI or MRI with serious motion artifact (n = 24); (d) incomplete pathological report or medical record (n = 14); (e) patients lost to follow-up (n = 20). Finally, a total of 174 patients diagnosed with EC were identified. The last follow-up was in March 2021.
2.2 MR Imaging
All patients underwent pelvic MRI and DWI before operation. MRI was performed on a 3.0T MRI system (Signa Excite HD, GE, Milwaukee, WI, USA) using an 8-channel body coil. Prior to MRI examinations, patients fasted for at least 4 hours and were given intramuscular 15 mg shyoscine butylbromide half an hour before examination. During image acquisition, the patient remained supine with a semi-filled bladder.
Scanning sequence and parameters: (1) Axial T1-weighted images (T1WI): a field of view (FOV): 38 cm × 26 cm, repetition time (TR): 500 ms, echo time (TE): 7.2 ms, slice thickness: 6 mm, inter-slice gap: 2 mm, matrix size: 352 × 192. (2) Axial and axial fast spin-echo (FSE) T2WI: FOV: 24 cm × 24 cm, TR: 4600 ms, TE: 68 ms, slice thickness: 3 mm, inter-slice gap: 1 mm, matrix size: 320 × 256; (3) Oblique sagittal T2WI: FOV: 26 cm × 24 cm, TR: 4600 ms, TE: 68 ms, slice thickness: 6 mm, inter-slice gap: 2 mm, matrix size: 320 × 256; (4) Axial DWI: FOV: 38 cm × 26 cm, TR: 4000 ms, TE: 65 ms, slice thickness: 4 mm, inter-slice gap: 1 mm, matrix size: 96 × 130, b-value: 0 and 1000 mm2/s.
2.3 Histologic and pathologic diagnosis
All surgical specimens were examined and reported by gynecologic pathologists. Tumor staging were performed according to standard 2018 FIGO criteria. Myometrial invasion , lymphovascular space invasion (LVSI) and lymph node metastasis (LNM) were confirmed under microscope according to the corresponding diagnostic criteria.
Postoperative follow-up of patients: 3 to 6 months for the first 2–3 years, 6 months until 5 years, and then annually, any time when there are related symptoms such as vaginal bleeding, abdominal distension and pain. Surveillance included gynecological examination, imaging and pathological biopsy if necessary.
2.4 Image Analysis
The images were assessed by two radiologists with 8 and 10 years (reader 1 and reader 2) of professional experience in pelvic MRI independently. Moreover, they were both blinded to each other’s results, and pathological and clinical data. Tumor was defined as a mass with signal hyperintensity on DWI and hypointensity on ADC map, compared with the signal of surrounding adjacent tissues. Meanwhile, T2WI sequence was referenced. ADC value measurements of the tumor were executed on an ADC map using GE Advantage Workstation 4.6, Function Tool software. The region of interest (ROI) was manually delineated in the slice containing the largest tumor layer, including the largest tumor area as much as possible and avoiding bleeding and necrotic tissue (Figure 1.A−C). Each reader measured ADC values for each patient three times, and the average value was taken.
Tumor segmentation were performed on axial T2WI sequence using ITK-SNAP (Version 3.6.0, http:// www.itksnap.org) software by reader 2. ROIs were sketched manually on all MRI levels containing tumor, and the ROIs covered the whole tumor as much as possible. Then the ROIs of all layers were fused to get the three-dimensional volume of interest (VOI). Finally, radiomic parameters were extract from VOIs using AK (Analysis Kit, Kinetics Version 2.1, GE Healthcare) software, following the IBSI standards, with a total of 1130 (Figure 1.D−E). To investigate the stability of radiomic features extracted by different readers and the same reader, 30 patients were randomly selected for tumor segmentation 2 weeks later.
2.5 Statistical Analysis
The data were analyzed with SPSS v. 26.0 (Chicago, IL, USA) and R (Version 3.6.1, http://www.r-project.org) .Patient clinicopathological characteristics were compared using a t-test (age, ADC and radscore) or chi-square test (FIGO stage, tumor size, pathological types, myometrial invasion, LVSI and LNM) . The intraclass correlation coefficient (ICC) was used to evaluate the intra-observer and inter-observer consistency of ADC values and radiomic parameters. The ADC values having higher intra-observer ICC was retained. The radiomic features with intra-observer and inter-observer ICC >0.75 were considered stable and were retained for subsequent analyses. Logistic regression analysis was used to select radiomic parameters. The radscore was calculated based on the regression coefficient of the selected radiomic features using multi-factor linear weighting. ADC value and radscore were compared by t-test. Receiver operating characteristic (ROC) curve and Youden index were used to obtain cut-off values of ADC vale and radscore. Spearman's bivariate correlation test was used to analyze the correlation between ADC value, radscore and recurrence. The Kaplan–Meier method was used to calculate survival rate and draw survival curve, Log–rank method was used for univariate analysis. Variables with P < 0.05 were included in a multivariate Cox regression model and independent prognostic factors of survival were identified with P<0.05.